Best pitcher RIGHT NOW

by phatj on Sep 17, 2010 1:55 AM EDT

Elsewhere today I asserted that Cole Hamels had been the best pitcher in the majors since the All-Star Break. I had no particular numbers to back this up, as BB-Ref doesn't have "split" leaderboards, so I thought about how to go about figuring it out. Then I wondered whether there was a way to estimate how good a pitcher is right at this moment. Season-long stats don't tell you this; they average out slumps, effects of nagging injuries, mastering new pitches, etc.

What I came up with is a weighted average, using the ordinal number of the pitcher's start as the weight - e.g. the pitcher's first game is weighted 1, the second is weighted 2, etc. This makes the pitcher's most recent starts much more important in the weighted average than the earliest starts, so if he's on a hot or cold streak it will reflect that. I think this would be a more accurate reflection of his true quality at any particular moment than looking at a short stretch of starts, as a hot or cold streak can reflect some luck and including the earlier starts in the average helps factor that out.

Putting this together was a major pain in the ass - I spent the better part of the last four hours on it. It required copying-and-pasting Hamels' game logs from BB-Ref into a spreadsheet, deleting the rows that don't contain data, then inserting columns and formulae for the calculated data (H/9, BB/9, etc. don't appear in the game logs) plus a whole set of additional columns for the weighted data. I started out with the logs for Halladay and Oswalt as well but got bogged down and don't have numbers to show at this time.

Without further ado, here are Hamels' numbers, both straight averages and weighted:

Straight Numbers

ERA

H/9

BB/9

K/9

K/BB

HR/9

GB/FB

6.49

3.00

7.86

2.54

9.29

3.65

1.11

0.91

Weighted Numbers

wIP

wERA

wH/9

wBB/9

wK/9

wK/BB

wHR/9

wGB/FB

6.68

2.44

7.12

2.86

9.53

4.21

0.83

0.93

As you can see, Hamels' numbers improve almost across the board, the only exception being his walk rate has trended up over the season, so his weighted walk rate is higher.

I wish I could show numbers for other pitchers. I'm going to have to pretty much start over to get this into a format where I can easily copy-and-paste logs and have it automatically do the rest; maybe next week I'll have time to do that. I can say that before I scrapped Halladay and Oswalt's numbers from my spreadsheet I was able to tell that Hamels' weighted ERA and K/9 were the best of the three; Halladay still had a big lead in BB/9 and K/BB though. Halladay's weighted ERA is higher than his straight ERA, while Oswalt's is lower.

Thoughts? Did I just waste a huge amount of time on this?

Tweet Comment 5 comments | 1 recs |

Do you like this story?

Comments

I love it.

Essentially, you’re trying to objectively measure a “power rankings” of sort for starting pitchers by more heavily weighing recent performances. The first thing I thought of was, “This is basically an attempt at an individual version of ”http://www.baseballprospectus.com/article.php?articleid=5247" target="new">ELO." I see definite value in that, even if the end result is bound to be imperfect.

I guess the main statistical question I’d have is what the coefficients should be to give the most accurate picture. The 1, 2, 3… 30 scale you’ve used makes sense intuitively, but I wonder if it provides the best statistical sample size to use. (I mean that as an open question, by the way, not as a criticism in any way.)

by PhillyFriar on Sep 17, 2010 8:26 AM EDT reply actions

Yes, objective power rankings is a great way to describe this.

I thought about the question of whether to use something other than ordinal weights but my math background is pretty far in my past at this point and never did include much in the way of statistics so I would be out of my depth trying to come up with better coefficients. If anyone who knows this kind of math better than me wants to take a stab at it, be my guest.

By the way, it occurs to me that this methodology, while devised for measuring pitchers in-season, could just as easily be applied to hitters, or to pitchers or hitters over a career. The latter could be used analogously (is that a word?) to something like a PECOTA projection to see how good a player really is at this point in his career. The downside to this method is that it requires a lot of data; where PECOTA just uses the aggregate lines for each season, this uses individual game logs. I guess that’s both a good thing (more data = more better) and a bad thing (more data = more work).

by phatj on Sep 17, 2010 8:47 AM EDT up reply actions

This is a good idea. Note by the way, that ESPN has pre- vs. post-ASG splits, and depending on what your IP cut off is, Hamels is at or near the top in ERA, WHIP, DIPS, and OPS Against (it doesn’t have FIP or xFIP).

by schmenkman on Sep 17, 2010 9:38 AM EDT reply actions